融合稠密连接网络与MLP-Mixer的频谱感知方法

A Spectrum Sensing Method Integrating DenseNet and MLP-Mixer

随着无线应用的持续激增,干扰环境下的通信变得愈发重要,频谱感知技术在解决无线电用频冲突方面发挥了重要作用。然而实际应用环境复杂,获取到的频谱信号不易被高效提取特征,这降低了频谱信号的实用性。如今人工智能在通信领域应用广泛,对通信技术产生重要影响。为此,从深度学习方法入手,提出一种融合稠密连接网络与MLP-Mixer的频谱感知方法。该模型首先通过Deepinsight网络对频谱信号数据实施处理与转换,使其变换为特征图像,再使用生成式对抗网络合成新的特征图,并在得到特征图像后,采用融合稠密连接网络的混合感知器提取特征,从而感知主用户信道占用情况。经过消融试验对比,所提方法相较于已有模型,较好地提升了频谱感知的检测概率。

Along with the surge of radio application,electronic communication in interference environments has become increasingly important.The spectrum sensing technique matters in surmounting the frequency conflict of radio.However,the complex environment hinders the efficient feature extraction from the received spectrum signal and reduces the signal practicality.Recently,the artificial intelligence has been widespread in communication field and crucially influenced the electronic countermeasures.Consequently,based on the deep learning,this work proposes a spectrum sensing method to mix DenseNet and MLP-Mixer.Firstly,the model processes and transforms the spectrum signal data to feature images by Deepinsight Net and the generative adversarial networks renew an image.After obtaining the feature image,aspectrum sensing method integrating DenseNet and MLP-Mixer is used in order to sense the channel occupancy of primary user.Compared with the existing model through ablation experiments,the proposed method improves the detection probability of spectrum sensing better.